Artificial Intelligence to Predict Pathology and Endoscopic Classification of Colorectal Polyps During Colonoscopy

What this Trial Is About

Researchers are evaluating an artificial intelligence (AI) model to predict the pathology and endoscopic classification of colorectal polyps during colonoscopy. Colonoscopy with optical diagnosis helps guide treatment and reduce unnecessary procedures, easing the burden on patients and healthcare systems. While current methods require extensive training and have limitations in accurately diagnosing all polyp types, AI-based computer-aided diagnosis (CADx) is rapidly developing and has shown promising accuracy for small lesions under 5mm. However, its effectiveness for larger polyps and serrated lesions, which are precancerous and more difficult to assess, remains unclear. The study involves developing an AI model using a large dataset of approximately 1600 cases, including serrated lesions, hyperplastic polyps, conventional adenomas, and early-stage colorectal cancer, collected retrospectively from pathology records. The model is built using popular AI classification algorithms and trained on static images linked to pathological diagnoses and endoscopic classifications. After optimization, the AI model's performance will be compared with that of endoscopists in a prospective cohort to assess its diagnostic accuracy. Participants are adults aged 18 years or older undergoing routine colonoscopy screening at multiple hospital centers who understand the study and provide consent. Researchers will collect data during colonoscopy and compare AI predictions to pathological outcomes. The main outcome measured is the accuracy of the AI optical diagnosis for colorectal polyps over two years. The study excludes individuals with certain serious health conditions, pregnancy, or prior colorectal surgery to ensure participant safety and data reliability.

AI in Predicting Polyp Pathology and Endoscopic Classification